Epipolar-Free 3D Gaussian Splatting for Generalizable Novel View Synthesis

Authors: Zhiyuan Min, Yawei Luo, Jianwen Sun, Yi Yang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate e Free Splat on wide-baseline novel view synthesis tasks using the Real Estate10K and ACID datasets. Extensive experiments demonstrate that e Free Splat surpasses state-of-the-art baselines that rely on epipolar priors, achieving superior geometry reconstruction and novel view synthesis quality.
Researcher Affiliation Academia Zhiyuan Min1 Yawei Luo1, Jianwen Sun2 Yi Yang1 1Zhejiang University 2Central China Normal University
Pseudocode No The paper does not contain a clearly labeled section or figure titled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Project page: https://tatakai1.github.io/efreesplat/.
Open Datasets Yes e Free Splat is trained on Real Estate10K [72] and ACID [26].
Dataset Splits No Following pixel Splat [6], we use the provided training and testing splits and evaluate three novel view images on each test scene. The paper does not explicitly mention a 'validation' split or its specific proportions/counts.
Hardware Specification Yes All models are trained on 4 RTX-4090 GPUs for 300, 000 iterations using the Adam optimizer [24].
Software Dependencies No The paper mentions using specific models like 'Vi T-B vision transformer' and 'Cro Co v2' and an 'Adam optimizer', but it does not specify version numbers for any software libraries or dependencies (e.g., PyTorch, TensorFlow, specific Python versions).
Experiment Setup Yes All models are trained on 4 RTX-4090 GPUs for 300, 000 iterations using the Adam optimizer [24]. The per-GPU batch size during training is 4. ... using the Adam optimizer with a learning rate 2e-4. ... the resolution of our training and testing images for fair comparison (256x256).